Documentation

Poll conversations.history for the channels you choose and store messages as memories – no webhook, no Socket Mode, no exposed endpoint.

Fetch issues + comments from the repos you list via the REST API – no webhook, no exposed endpoint.

Import a local Beads project via the bd CLI, deduped by id, run on a timer.

Scan a vault and store notes as memories – runs where the vault lives, writes to your configured backend.

Map Ogham’s compile_wiki, topic_summaries, and TLDR resolutions onto the consolidation pattern.

Three things Ogham deliberately doesn’t do.

Where the name comes from and what the Irish terms in the codebase mean

Microsoft Agent Framework GA’d 1.0 in Q1 2026 as the merged successor to Semantic Kernel and AutoGen. It ships in Python and .NET with pluggable memory: built-in backends include Foundry Agent Service Memory, Mem0, Redis, and Neo4j. It also speaks MCP natively, which means Ogham can plug in today with zero glue code.

This page covers Shape A: registering Ogham as an MCP tool. The agent decides when to call it – store, search, compile, walk the graph – the same way it would call any other tool. No subclassing, no provider class.

What is SCION? #

SCION is Google Cloud’s multi-agent orchestration testbed. It runs LLM agents (Gemini CLI, Claude Code, OpenCode, Codex) in isolated Docker or Kubernetes containers, each with their own identity, credentials, and workspace.

The problem: agents in SCION are isolated by design. Agent A doesn’t know what Agent B learned. Every agent starts from a blank slate, and when it stops, its knowledge stays locked in its container.